2 * A Simple Sandbox for MO-Eval
4 * (c) 2001--2008 Martin Mares <mj@ucw.cz>
7 #define _LARGEFILE64_SOURCE
22 #include <sys/ptrace.h>
23 #include <sys/signal.h>
24 #include <sys/sysinfo.h>
25 #include <sys/syscall.h>
26 #include <sys/resource.h>
28 #define NONRET __attribute__((noreturn))
29 #define UNUSED __attribute__((unused))
30 #define ARRAY_SIZE(a) (int)(sizeof(a)/sizeof(a[0]))
32 static int filter_syscalls; /* 0=off, 1=liberal, 2=totalitarian */
33 static int timeout; /* milliseconds */
34 static int wall_timeout;
35 static int pass_environ;
36 static int file_access;
38 static int memory_limit;
39 static char *redir_stdin, *redir_stdout, *redir_stderr;
43 static int is_ptraced;
44 static volatile int timer_tick;
45 static struct timeval start_time;
46 static int ticks_per_sec;
48 static int partial_line;
50 static void die(char *msg, ...) NONRET;
54 static FILE *metafile;
57 meta_open(const char *name)
59 if (!strcmp(name, "-"))
64 metafile = fopen(name, "w");
66 die("Failed to open metafile '%s'",name);
72 if (metafile && metafile != stdout)
76 static void __attribute__((format(printf,1,2)))
77 meta_printf(const char *fmt, ...)
84 vfprintf(metafile, fmt, args);
88 static int total_ms, wall_ms;
91 final_stats(struct rusage *rus)
93 struct timeval total, now, wall;
94 timeradd(&rus->ru_utime, &rus->ru_stime, &total);
95 total_ms = total.tv_sec*1000 + total.tv_usec/1000;
96 gettimeofday(&now, NULL);
97 // FIXME: We are not guaranteed to have start_time always initialized
98 timersub(&now, &start_time, &wall);
99 wall_ms = wall.tv_sec*1000 + wall.tv_usec/1000;
101 meta_printf("time:%d.%03d\n", total_ms/1000, total_ms%1000);
102 meta_printf("time-wall:%d.%03d\n", wall_ms/1000, wall_ms%1000);
105 /*** Messages and exits ***/
113 ptrace(PTRACE_KILL, box_pid);
114 kill(-box_pid, SIGKILL);
115 kill(box_pid, SIGKILL);
119 int p = wait4(box_pid, &stat, 0, &rus);
121 fprintf(stderr, "UGH: Lost track of the process\n");
137 static void NONRET __attribute__((format(printf,1,2)))
143 vfprintf(stderr, msg, args);
148 static void NONRET __attribute__((format(printf,1,2)))
154 vfprintf(stderr, msg, args);
159 static void __attribute__((format(printf,1,2)))
166 int len = strlen(msg);
168 partial_line = (msg[len-1] != '\n');
169 vfprintf(stderr, msg, args);
178 void *p = malloc(size);
180 die("Out of memory");
184 /*** Syscall rules ***/
186 static const char * const syscall_names[] = {
187 #include "box/syscall-table.h"
189 #define NUM_SYSCALLS ARRAY_SIZE(syscall_names)
190 #define NUM_ACTIONS (NUM_SYSCALLS+64)
193 A_DEFAULT, // Use the default action
194 A_NO, // Always forbid
195 A_YES, // Always permit
196 A_FILENAME, // Permit if arg1 is a known filename
197 A_LIBERAL = 128, // Valid only in liberal mode
200 static unsigned char syscall_action[NUM_ACTIONS] = {
201 #define S(x) [__NR_##x]
203 // Syscalls permitted for specific file names
204 S(open) = A_FILENAME,
205 S(creat) = A_FILENAME,
206 S(unlink) = A_FILENAME,
207 S(oldstat) = A_FILENAME,
208 S(access) = A_FILENAME,
209 S(oldlstat) = A_FILENAME,
210 S(truncate) = A_FILENAME,
211 S(stat) = A_FILENAME,
212 S(lstat) = A_FILENAME,
213 S(truncate64) = A_FILENAME,
214 S(stat64) = A_FILENAME,
215 S(lstat64) = A_FILENAME,
216 S(readlink) = A_FILENAME,
218 // Syscalls permitted always
233 S(ftruncate) = A_YES,
235 S(personality) = A_YES,
239 S(getresuid) = A_YES,
247 S(ftruncate64) = A_YES,
256 S(set_thread_area) = A_YES,
257 S(get_thread_area) = A_YES,
258 S(exit_group) = A_YES,
260 // Syscalls permitted only in liberal mode
261 S(time) = A_YES | A_LIBERAL,
262 S(alarm) = A_YES | A_LIBERAL,
263 S(pause) = A_YES | A_LIBERAL,
264 S(signal) = A_YES | A_LIBERAL,
265 S(fchmod) = A_YES | A_LIBERAL,
266 S(sigaction) = A_YES | A_LIBERAL,
267 S(sgetmask) = A_YES | A_LIBERAL,
268 S(ssetmask) = A_YES | A_LIBERAL,
269 S(sigsuspend) = A_YES | A_LIBERAL,
270 S(sigpending) = A_YES | A_LIBERAL,
271 S(getrlimit) = A_YES | A_LIBERAL,
272 S(getrusage) = A_YES | A_LIBERAL,
273 S(ugetrlimit) = A_YES | A_LIBERAL,
274 S(gettimeofday) = A_YES | A_LIBERAL,
275 S(select) = A_YES | A_LIBERAL,
276 S(readdir) = A_YES | A_LIBERAL,
277 S(setitimer) = A_YES | A_LIBERAL,
278 S(getitimer) = A_YES | A_LIBERAL,
279 S(sigreturn) = A_YES | A_LIBERAL,
280 S(mprotect) = A_YES | A_LIBERAL,
281 S(sigprocmask) = A_YES | A_LIBERAL,
282 S(getdents) = A_YES | A_LIBERAL,
283 S(getdents64) = A_YES | A_LIBERAL,
284 S(_newselect) = A_YES | A_LIBERAL,
285 S(fdatasync) = A_YES | A_LIBERAL,
286 S(mremap) = A_YES | A_LIBERAL,
287 S(poll) = A_YES | A_LIBERAL,
288 S(getcwd) = A_YES | A_LIBERAL,
289 S(nanosleep) = A_YES | A_LIBERAL,
290 S(rt_sigreturn) = A_YES | A_LIBERAL,
291 S(rt_sigaction) = A_YES | A_LIBERAL,
292 S(rt_sigprocmask) = A_YES | A_LIBERAL,
293 S(rt_sigpending) = A_YES | A_LIBERAL,
294 S(rt_sigtimedwait) = A_YES | A_LIBERAL,
295 S(rt_sigqueueinfo) = A_YES | A_LIBERAL,
296 S(rt_sigsuspend) = A_YES | A_LIBERAL,
297 S(mmap2) = A_YES | A_LIBERAL,
298 S(_sysctl) = A_YES | A_LIBERAL,
303 syscall_name(unsigned int id, char *buf)
305 if (id < NUM_SYSCALLS && syscall_names[id])
306 return syscall_names[id];
309 sprintf(buf, "#%d", id);
315 syscall_by_name(char *name)
317 for (unsigned int i=0; i<NUM_SYSCALLS; i++)
318 if (syscall_names[i] && !strcmp(syscall_names[i], name))
325 unsigned long l = strtoul(name, &ep, 0);
328 if (l >= NUM_ACTIONS)
334 set_syscall_action(char *a)
336 char *sep = strchr(a, '=');
337 enum action act = A_YES;
341 if (!strcmp(sep, "yes"))
343 else if (!strcmp(sep, "no"))
345 else if (!strcmp(sep, "file"))
351 int sys = syscall_by_name(a);
353 die("Unknown syscall `%s'", a);
354 if (sys >= NUM_ACTIONS)
355 die("Syscall `%s' out of range", a);
356 syscall_action[sys] = act;
365 struct path_rule *next;
368 static struct path_rule default_path_rules[] = {
371 { "/usr/lib/", A_YES },
372 { "/opt/lib/", A_YES },
373 { "/usr/share/zoneinfo/", A_YES },
374 { "/usr/share/locale/", A_YES },
375 { "/dev/null", A_YES },
376 { "/dev/zero", A_YES },
377 { "/proc/meminfo", A_YES },
378 { "/proc/self/stat", A_YES },
379 { "/proc/self/exe", A_YES }, // Needed by FPC 2.0.x runtime
382 static struct path_rule *user_path_rules;
383 static struct path_rule **last_path_rule = &user_path_rules;
386 set_path_action(char *a)
388 char *sep = strchr(a, '=');
389 enum action act = A_YES;
393 if (!strcmp(sep, "yes"))
395 else if (!strcmp(sep, "no"))
401 struct path_rule *r = xmalloc(sizeof(*r) + strlen(a) + 1);
402 r->path = (char *)(r+1);
407 last_path_rule = &r->next;
412 match_path_rule(struct path_rule *r, char *path)
416 if (*rr++ != *path++)
418 if (rr[-1] == '/' && !path[-1])
422 if (rr > r->path && rr[-1] != '/' && *path)
427 /*** Environment rules ***/
430 char *var; // Variable to match
431 char *val; // ""=clear, NULL=inherit
433 struct env_rule *next;
436 static struct env_rule *first_env_rule;
437 static struct env_rule **last_env_rule = &first_env_rule;
439 static struct env_rule default_env_rules[] = {
440 { "LIBC_FATAL_STDERR_", "1" }
444 set_env_action(char *a0)
446 struct env_rule *r = xmalloc(sizeof(*r) + strlen(a0) + 1);
447 char *a = (char *)(r+1);
450 char *sep = strchr(a, '=');
462 last_env_rule = &r->next;
468 match_env_var(char *env_entry, struct env_rule *r)
470 if (strncmp(env_entry, r->var, r->var_len))
472 return (env_entry[r->var_len] == '=');
476 apply_env_rule(char **env, int *env_sizep, struct env_rule *r)
478 // First remove the variable if already set
480 while (pos < *env_sizep && !match_env_var(env[pos], r))
482 if (pos < *env_sizep)
485 env[pos] = env[*env_sizep];
486 env[*env_sizep] = NULL;
489 // What is the new value?
495 new = xmalloc(r->var_len + 1 + strlen(r->val) + 1);
496 sprintf(new, "%s=%s", r->var, r->val);
501 while (environ[pos] && !match_env_var(environ[pos], r))
503 if (!(new = environ[pos]))
507 // Add it at the end of the array
508 env[(*env_sizep)++] = new;
509 env[*env_sizep] = NULL;
513 setup_environment(void)
515 // Link built-in rules with user rules
516 for (int i=ARRAY_SIZE(default_env_rules)-1; i >= 0; i--)
518 default_env_rules[i].next = first_env_rule;
519 first_env_rule = &default_env_rules[i];
522 // Scan the original environment
523 char **orig_env = environ;
525 while (orig_env[orig_size])
528 // For each rule, reserve one more slot and calculate length
530 for (struct env_rule *r = first_env_rule; r; r=r->next)
533 r->var_len = strlen(r->var);
536 // Create a new environment
537 char **env = xmalloc((orig_size + num_rules + 1) * sizeof(char *));
541 memcpy(env, environ, orig_size * sizeof(char *));
548 // Apply the rules one by one
549 for (struct env_rule *r = first_env_rule; r; r=r->next)
550 apply_env_rule(env, &size, r);
552 // Return the new env and pass some gossip
555 fprintf(stderr, "Passing environment:\n");
556 for (int i=0; env[i]; i++)
557 fprintf(stderr, "\t%s\n", env[i]);
562 /*** Syscall checks ***/
565 valid_filename(unsigned long addr)
567 char namebuf[4096], *p, *end;
571 err("File access forbidden");
572 if (file_access >= 9)
577 sprintf(namebuf, "/proc/%d/mem", (int) box_pid);
578 mem_fd = open(namebuf, O_RDONLY);
580 die("open(%s): %m", namebuf);
587 int remains = PAGE_SIZE - (addr & (PAGE_SIZE-1));
588 int l = namebuf + sizeof(namebuf) - end;
592 err("Access to file with name too long");
593 if (lseek64(mem_fd, addr, SEEK_SET) < 0)
594 die("lseek64(mem): %m");
595 remains = read(mem_fd, end, l);
597 die("read(mem): %m");
599 err("Access to file with name out of memory");
606 msg("[%s] ", namebuf);
607 if (file_access >= 3)
610 // Everything in current directory is permitted
611 if (!strchr(namebuf, '/') && strcmp(namebuf, ".."))
614 // ".." anywhere in the path is forbidden
615 enum action act = A_DEFAULT;
616 if (strstr(namebuf, ".."))
620 for (struct path_rule *r = user_path_rules; r && !act; r=r->next)
621 act = match_path_rule(r, namebuf);
623 // Scan built-in rules
624 if (file_access >= 2)
625 for (int i=0; i<ARRAY_SIZE(default_path_rules) && !act; i++)
626 act = match_path_rule(&default_path_rules[i], namebuf);
629 err("Forbidden access to file `%s'", namebuf);
633 valid_syscall(struct user *u)
635 unsigned int sys = u->regs.orig_eax;
636 enum action act = (sys < NUM_ACTIONS) ? syscall_action[sys] : A_DEFAULT;
640 if (filter_syscalls == 1)
652 valid_filename(u->regs.ebx);
660 if (u->regs.ebx == box_pid)
661 err("Committed suicide by signal %d", (int)u->regs.ecx);
664 if (u->regs.ebx == box_pid && u->regs.ecx == box_pid)
665 err("Committed suicide by signal %d", (int)u->regs.edx);
673 signal_alarm(int unused UNUSED)
675 /* Time limit checks are synchronous, so we only schedule them there. */
681 signal_int(int unused UNUSED)
683 /* Interrupts are fatal, so no synchronization requirements. */
692 struct timeval now, wall;
694 gettimeofday(&now, NULL);
695 timersub(&now, &start_time, &wall);
696 wall_ms = wall.tv_sec*1000 + wall.tv_usec/1000;
697 if (wall_ms > wall_timeout)
698 err("Time limit exceeded (wall clock)");
700 fprintf(stderr, "[wall time check: %d msec]\n", wall_ms);
705 int c, utime, stime, ms;
706 static int proc_status_fd;
709 sprintf(buf, "/proc/%d/stat", (int) box_pid);
710 proc_status_fd = open(buf, O_RDONLY);
711 if (proc_status_fd < 0)
712 die("open(%s): %m", buf);
714 lseek(proc_status_fd, 0, SEEK_SET);
715 if ((c = read(proc_status_fd, buf, sizeof(buf)-1)) < 0)
716 die("read on /proc/$pid/stat: %m");
717 if (c >= (int) sizeof(buf) - 1)
718 die("/proc/$pid/stat too long");
721 while (*x && *x != ' ')
726 die("proc syntax error 1");
727 while (*x && (*x != ')' || x[1] != ' '))
729 while (*x == ')' || *x == ' ')
731 if (sscanf(x, "%*c %*d %*d %*d %*d %*d %*d %*d %*d %*d %*d %d %d", &utime, &stime) != 2)
732 die("proc syntax error 2");
733 ms = (utime + stime) * 1000 / ticks_per_sec;
735 fprintf(stderr, "[time check: %d msec]\n", ms);
737 err("Time limit exceeded");
744 int syscall_count = 0;
748 bzero(&sa, sizeof(sa));
749 sa.sa_handler = signal_int;
750 sigaction(SIGINT, &sa, NULL);
751 gettimeofday(&start_time, NULL);
752 ticks_per_sec = sysconf(_SC_CLK_TCK);
753 if (ticks_per_sec <= 0)
754 die("Invalid ticks_per_sec!");
755 if (timeout || wall_timeout)
757 sa.sa_handler = signal_alarm;
758 sigaction(SIGALRM, &sa, NULL);
771 p = wait4(box_pid, &stat, WUNTRACED, &rus);
779 die("wait4: unknown pid %d exited!", p);
784 // FIXME: If the process has exited before being ptraced, signal an internal error
785 if (WEXITSTATUS(stat))
786 err("Exited with error status %d", WEXITSTATUS(stat));
787 if (timeout && total_ms > timeout)
788 err("Time limit exceeded");
789 if (wall_timeout && wall_ms > wall_timeout)
790 err("Time limit exceeded (wall clock)");
792 fprintf(stderr, "OK (%d.%03d sec real, %d.%03d sec wall, %d syscalls)\n",
793 total_ms/1000, total_ms%1000,
794 wall_ms/1000, wall_ms%1000,
798 if (WIFSIGNALED(stat))
801 err("Caught fatal signal %d%s", WTERMSIG(stat), (syscall_count ? "" : " during startup"));
803 if (WIFSTOPPED(stat))
805 int sig = WSTOPSIG(stat);
809 static int stop_count = -1;
810 if (ptrace(PTRACE_GETREGS, box_pid, NULL, &u) < 0)
811 die("ptrace(PTRACE_GETREGS): %m");
813 if (!stop_count) /* Traceme request */
814 msg(">> Traceme request caught\n");
815 else if (stop_count & 1) /* Syscall entry */
818 msg(">> Syscall %-12s (%08lx,%08lx,%08lx) ", syscall_name(u.regs.orig_eax, namebuf), u.regs.ebx, u.regs.ecx, u.regs.edx);
822 if (u.regs.orig_eax == __NR_execve)
825 else if (valid_syscall(&u))
830 * Unfortunately, PTRACE_KILL kills _after_ the syscall completes,
831 * so we have to change it to something harmless (e.g., an undefined
832 * syscall) and make the program continue.
834 unsigned int sys = u.regs.orig_eax;
835 u.regs.orig_eax = 0xffffffff;
836 if (ptrace(PTRACE_SETREGS, box_pid, NULL, &u) < 0)
837 die("ptrace(PTRACE_SETREGS): %m");
838 err("Forbidden syscall %s", syscall_name(sys, namebuf));
841 else /* Syscall return */
842 msg("= %ld\n", u.regs.eax);
843 ptrace(PTRACE_SYSCALL, box_pid, 0, 0);
845 else if (sig != SIGSTOP && sig != SIGXCPU && sig != SIGXFSZ)
847 msg(">> Signal %d\n", sig);
848 ptrace(PTRACE_SYSCALL, box_pid, 0, sig);
851 err("Received signal %d", sig);
854 die("wait4: unknown status %x, giving up!", stat);
859 box_inside(int argc, char **argv)
864 memcpy(args, argv, argc * sizeof(char *));
866 if (set_cwd && chdir(set_cwd))
871 if (open(redir_stdin, O_RDONLY) != 0)
872 die("open(\"%s\"): %m", redir_stdin);
877 if (open(redir_stdout, O_WRONLY | O_CREAT | O_TRUNC, 0666) != 1)
878 die("open(\"%s\"): %m", redir_stdout);
883 if (open(redir_stderr, O_WRONLY | O_CREAT | O_TRUNC, 0666) != 2)
884 die("open(\"%s\"): %m", redir_stderr);
891 rl.rlim_cur = rl.rlim_max = memory_limit * 1024;
892 if (setrlimit(RLIMIT_AS, &rl) < 0)
893 die("setrlimit: %m");
895 rl.rlim_cur = rl.rlim_max = 64;
896 if (setrlimit(RLIMIT_NOFILE, &rl) < 0)
897 die("setrlimit: %m");
900 if (ptrace(PTRACE_TRACEME) < 0)
901 die("ptrace(PTRACE_TRACEME): %m");
902 /* Trick: Make sure that we are stopped until the boxkeeper wakes up. */
903 signal(SIGCHLD, SIG_IGN);
906 execve(args[0], args, setup_environment());
907 die("execve(\"%s\"): %m", args[0]);
913 fprintf(stderr, "Invalid arguments!\n");
915 Usage: box [<options>] -- <command> <arguments>\n\
918 -a <level>\tSet file access level (0=none, 1=cwd, 2=/etc,/lib,..., 3=whole fs, 9=no checks; needs -f)\n\
919 -c <dir>\tChange directory to <dir> first\n\
920 -e\t\tInherit full environment of the parent process\n\
921 -E <var>\tInherit the environment variable <var> from the parent process\n\
922 -E <var>=<val>\tSet the environment variable <var> to <val>; unset it if <var> is empty\n\
923 -f\t\tFilter system calls (-ff=very restricted)\n\
924 -i <file>\tRedirect stdin from <file>\n\
925 -m <size>\tLimit address space to <size> KB\n\
926 -M <file>\tOutput process information to <file> (name:value)\n\
927 -o <file>\tRedirect stdout to <file>\n\
928 -p <path>\tPermit access to the specified path (or subtree if it ends with a `/')\n\
929 -p <path>=<act>\tDefine action for the specified path (<act>=yes/no)\n\
930 -r <file>\tRedirect stderr to <file>\n\
931 -s <sys>\tPermit the specified syscall (be careful)\n\
932 -s <sys>=<act>\tDefine action for the specified syscall (<act>=yes/no/file)\n\
933 -t <time>\tSet run time limit (seconds, fractions allowed)\n\
934 -T\t\tAllow syscalls for measuring run time\n\
935 -v\t\tBe verbose (use multiple times for even more verbosity)\n\
936 -w <time>\tSet wall clock time limit (seconds, fractions allowed)\n\
942 main(int argc, char **argv)
947 while ((c = getopt(argc, argv, "a:c:eE:fi:m:M:o:p:r:s:t:Tvw:")) >= 0)
951 file_access = atol(optarg);
960 if (!set_env_action(optarg))
967 redir_stdin = optarg;
970 memory_limit = atol(optarg);
976 redir_stdout = optarg;
979 if (!set_path_action(optarg))
983 redir_stderr = optarg;
986 if (!set_syscall_action(optarg))
990 timeout = 1000*atof(optarg);
993 syscall_action[__NR_times] = A_YES;
999 wall_timeout = 1000*atof(optarg);
1008 if (setreuid(uid, uid) < 0)
1009 die("setreuid: %m");
1014 box_inside(argc-optind, argv+optind);
1017 die("Internal error: fell over edge of the world");